Tech Desk: OpenAI has ignited a firestorm in the scientific community, claiming an internal AI model “more capable than GPT-6 Astra” has cracked a 90-year-old, $1 million math mystery: proving the three-dimensional Navier-Stokes equations can develop a finite-time singularity. This monumental achievement, if confirmed, would resolve one of the seven Millennium Prize Problems, with OpenAI stating it involved 10,000 parallel agents and was verified using the programming language Lean.

However, the announcement has been met with significant controversy. Just before OpenAI’s reveal, mathematicians Tristan Buckmaster and Levent Alpoge independently posted a related result, having worked on the problem for months, also utilizing AI models. Buckmaster subsequently raised serious questions, alleging OpenAI might have accessed their private drafts stored in its Codex tool and that an OpenAI researcher pressured Alpoge regarding authorship due to his Anthropic employment. OpenAI has vehemently denied these claims, asserting it did not access specific user data and that discussions followed standard academic norms.

Fields Medalist Terence Tao, a leading mathematician, voiced broader concerns, warning of a “chilling effect” on the field. He fears AI is being used to produce answers without the crucial accompanying insight into “why” an approach works, which is often more valuable for scientific progress. Tao also highlighted that AI companies rarely disclose failed attempts, potentially hindering future breakthroughs and discouraging mathematicians from sharing unfinished work. This unfolding drama underscores the complex challenges and ethical dilemmas emerging at the intersection of AI and fundamental scientific discovery.